Automated detection of mouse scratching behaviour using convolutional spiking neural network
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Complex adaptive systems (MPCAS), MSc
Publicerad
2023
Författare
Ma, Kelly
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
In recent years, significant progress has been achieved in the field of artificial neural
networks (ANNs). However, their performance still falls short when compared to
their biological counterparts. As a result, there is growing interest in Spiking neural
networks (SNNs), which are known for their biological plausibility and efficiency,
aiming to combine the advantages of both existing ANNs and biological networks.
Therefore, this project focuses on the development of an SNN that can reliably
identify mouse scratching behaviour. The test animals were recorded using an event
camera, and after the data was preprocessed and converted into frames, a data
set was created. Among the different training methods tested, the SNN trained
with backpropagation combined with the gradient surrogate method yielded the
best result, achieving a final accuracy of approximately 97% on the test dataset.
However, it should also be noted that the preprocessing process actually has a
significant impact on overall performance. Taking this into consideration, the actual
accuracy would be around 83%. This result obtained indicate that a SNN is capable
of detecting scratching behaviour despite the limitations imposed by the small size
of the available data, thus showcasing the potential of SNNs in real-life applications
Beskrivning
Ämne/nyckelord
ANN, SNN, dynamic behaviour detection